A0365
Title: Predicting patient trajectories with deep multi-state models trained on electronic health record data
Authors: Thomas Matcham - Imperial College London (United Kingdom) [presenting]
Abstract: In recent years, a range of survival models incorporating deep learning elements have been produced to better model survival data, particularly when very large quantities of data are available. Electronic health records contain enormous quantities of patient trajectory data, giving time-to-event data for the progression of diseases as well as consecutive interactions with healthcare providers. The extension of deep learning survival models is explored in the multi-state setting, modelling long-term COVID-19 hospitalization outcomes and the progression of type II diabetes-related chronic diseases.
